Selected work Case study 04
Making a powerful conversational AI platform usable
Conversational AI · Platform design
Two sides of the conversation
The problem
01Problem
Koopid could automate customer conversations and hand off to live agents, but the screens were organized around how the platform was built rather than the decisions an administrator, supervisor, or agent was trying to make.
02Why it mattered
Conversation state, routing, and escalation were rarely visible, so daily users spent attention reconstructing context instead of resolving the customer in front of them.
03Solution
Customer automation, the agent desktop, and the admin tools treated as a single service journey, with visible status, role-shaped navigation, and reusable patterns across all three.
Case overview
Koopid could automate customer conversations and support live-agent escalation, but the interface reflected the complexity of the underlying technology. As the product designer, I created a clearer system across customer conversations, the agent desktop, and administrative tools.
Ground-level research
I reviewed the product with customer success, engineers, and the people using each role. Three signals kept repeating: customers escalated from automation too early, agents had to interpret unclear states, and interface confusion created avoidable support work.
01 · Interface audit
The technical capability was present, but hierarchy and status were difficult to scan across the chatbot, agent desktop, and portal.
02 · Service journey
Mapping the complete journey showed how customer-side friction became downstream work for agents and support teams.
Users and decisions
Onboarding, configuration, branding, and workflow creation
Administrators needed enough control to configure a customer instance without understanding the platform like the team that built it.
Agent oversight, sentiment review, and conflict resolution
Supervisors used conversation patterns and agent feedback to improve automation while keeping escalations under control.
Customer conversations and escalation handling
Agents were the primary daily users. Their experience connected the quality of the customer conversation to the quality of the operational tools behind it.
Product direction
The goal was not to make the product look simpler. It was to help each person understand the current state, the next action, and the consequence of that action without absorbing the platform's internal complexity.
Clear conversation state, routing, and escalation cues reduced the need to infer what the system was doing.
Navigation and permissions reflected the different decisions an administrator, supervisor, or agent needed to make.
Customer-facing automation and agent tools were treated as one service journey, not unrelated interfaces.
Shared components established consistency across the agent desktop, admin portal, and future product work.
Shared hierarchy makes configuration predictable across roles.
Workflow setup is organized around clear tasks, state, and feedback.
Collaboration and delivery
As the only product designer, I could not become a review gate for every decision. I established reusable patterns, clarified what needed live collaboration, and documented what engineers could carry forward independently. Reviews focused on user impact, technical constraints, and unresolved risk.
I brought structure to ideas arriving from customers, engineering, and leadership, then made the tradeoffs explicit.
Short feedback loops across time zones kept decisions moving. The unglamorous work of documenting states and ownership prevented avoidable rework.
Reusable interaction and component patterns improved the next feature, not only the screen being redesigned.
Outcome
Public counts are not available, so the evidence is presented in the form the team could verify.
The product shipped and supported enterprise contact-center deployments, including Avaya.
Customer success reported substantially fewer support tickets and configuration errors, while users praised the more intuitive and streamlined workflows.
Reusable React components were used throughout the app. The componentized front end was adaptable by developers and designed to support white-label deployments.
Strong adoption, usability, and product capability strengthened Koopid's position before Dialpad acquired the company in early 2021. The redesign contributed to that product readiness, but it was not the sole cause of the acquisition.
Reflection
A choppy automated conversation did not end with the customer. It changed the work arriving at the agent desktop, the supervisor's workload, and the support team's queue. Treating that chain as one experience led to better priorities.
If I repeated the project, I would instrument escalation reasons and time-to-resolution sooner. The qualitative signal was strong, but earlier operational baselines would make the impact easier to isolate.
